s390/pci: Sort PCI functions prior to creating virtual busses

Instead of relying on the observed but not architected firmware behavior
that PCI functions from the same card are listed in ascending RID order
in clp_list_pci() ensure this by sorting. To allow for sorting separate
the initial clp_list_pci() and creation of the virtual PCI busses.

Note that fundamentally in our per-PCI function hotplug design non RID
order of discovery is still possible. For example when the two PFs of
a two port NIC are hotplugged after initial boot and in descending RID
order. In this case the virtual PCI bus would be created by the second
PF using that PF's UID as domain number instead of that of the first PF.
Thus the domain number would then change from the UID of the second PF
to that of the first PF on reboot but there is really nothing we can do
about that since changing domain numbers at runtime seems even worse.
This only impacts the domain number as the RIDs are consistent and thus
even with just the second PF visible it will show up in the correct
position on the virtual bus.

Reviewed-by: Gerd Bayer <[email protected]>
Signed-off-by: Niklas Schnelle <[email protected]>
Signed-off-by: Heiko Carstens <[email protected]>

1 files changed, 61 insertions, 8 deletions

diff --git a/arch/s390/pci/pci.c b/arch/s390/pci/pci.c
index bd9624c20b80..b7efa96776ea 100644
--- a/arch/s390/pci/pci.c
+++ b/arch/s390/pci/pci.c
@@ -29,6 +29,7 @@
 #include <linux/pci.h>
 #include <linux/printk.h>
 #include <linux/lockdep.h>
+#include <linux/list_sort.h>
 
 #include <asm/isc.h>
 #include <asm/airq.h>
@@ -785,7 +786,6 @@ struct zpci_dev *zpci_create_device(u32 fid, u32 fh, enum zpci_state state)
 	struct zpci_dev *zdev;
 	int rc;
 
-	zpci_dbg(1, "add fid:%x, fh:%x, c:%d\n", fid, fh, state);
 	zdev = kzalloc(sizeof(*zdev), GFP_KERNEL);
 	if (!zdev)
 		return ERR_PTR(-ENOMEM);
@@ -805,6 +805,19 @@ struct zpci_dev *zpci_create_device(u32 fid, u32 fh, enum zpci_state state)
 	mutex_init(&zdev->fmb_lock);
 	mutex_init(&zdev->kzdev_lock);
 
+	return zdev;
+
+error:
+	zpci_dbg(0, "crt fid:%x, rc:%d\n", fid, rc);
+	kfree(zdev);
+	return ERR_PTR(rc);
+}
+
+int zpci_add_device(struct zpci_dev *zdev)
+{
+	int rc;
+
+	zpci_dbg(1, "add fid:%x, fh:%x, c:%d\n", zdev->fid, zdev->fh, zdev->state);
 	rc = zpci_init_iommu(zdev);
 	if (rc)
 		goto error;
@@ -816,15 +829,13 @@ struct zpci_dev *zpci_create_device(u32 fid, u32 fh, enum zpci_state state)
 	spin_lock(&zpci_list_lock);
 	list_add_tail(&zdev->entry, &zpci_list);
 	spin_unlock(&zpci_list_lock);
-
-	return zdev;
+	return 0;
 
 error_destroy_iommu:
 	zpci_destroy_iommu(zdev);
 error:
-	zpci_dbg(0, "add fid:%x, rc:%d\n", fid, rc);
-	kfree(zdev);
-	return ERR_PTR(rc);
+	zpci_dbg(0, "add fid:%x, rc:%d\n", zdev->fid, rc);
+	return rc;
 }
 
 bool zpci_is_device_configured(struct zpci_dev *zdev)
@@ -1082,6 +1093,49 @@ bool zpci_is_enabled(void)
 	return s390_pci_initialized;
 }
 
+static int zpci_cmp_rid(void *priv, const struct list_head *a,
+			const struct list_head *b)
+{
+	struct zpci_dev *za = container_of(a, struct zpci_dev, entry);
+	struct zpci_dev *zb = container_of(b, struct zpci_dev, entry);
+
+	/*
+	 * PCI functions without RID available maintain original order
+	 * between themselves but sort before those with RID.
+	 */
+	if (za->rid == zb->rid)
+		return za->rid_available > zb->rid_available;
+	/*
+	 * PCI functions with RID sort by RID ascending.
+	 */
+	return za->rid > zb->rid;
+}
+
+static void zpci_add_devices(struct list_head *scan_list)
+{
+	struct zpci_dev *zdev, *tmp;
+
+	list_sort(NULL, scan_list, &zpci_cmp_rid);
+	list_for_each_entry_safe(zdev, tmp, scan_list, entry) {
+		list_del_init(&zdev->entry);
+		zpci_add_device(zdev);
+	}
+}
+
+int zpci_scan_devices(void)
+{
+	LIST_HEAD(scan_list);
+	int rc;
+
+	rc = clp_scan_pci_devices(&scan_list);
+	if (rc)
+		return rc;
+
+	zpci_add_devices(&scan_list);
+	zpci_bus_scan_busses();
+	return 0;
+}
+
 static int __init pci_base_init(void)
 {
 	int rc;
@@ -1111,10 +1165,9 @@ static int __init pci_base_init(void)
 	if (rc)
 		goto out_irq;
 
-	rc = clp_scan_pci_devices();
+	rc = zpci_scan_devices();
 	if (rc)
 		goto out_find;
-	zpci_bus_scan_busses();
 
 	s390_pci_initialized = 1;
 	return 0;